Partial Discharge Pattern Recognition of Transformers Based on MobileNets Convolutional Neural Network

Abstract

The power system on the offshore platform is of great importance since it is the power source for oil and gas exploitation, procession and transportation. Transformers constitute key equipment in the power system, and partial discharge (PD) is its most common fault that should be monitored and identified ın a timely and accurate manner. However, the existing PD classifiers cannot meet the demand for real-time online monitoring due to their disadvantages of high memory consumption and poor timeliness. Therefore, a new MobileNets convolutional neural network (MCNN) model is proposed to identify the PD pattern of transformers based on the phase resolved partial discharge (PRPD) spectrum. The model has the advantages of low computational complexity, fast reasoning speed and excellent classification performance. Firstly, we make four typical defect models of PD and conduct a test in a laboratory to collect the PRPD spectra as the data sample. In order to further improve the feature expression ability and recognition accuracy of the model, the lightweight attention mechanism Squeeze-and-Excitation (SE) module and the nonlinear function hard-swish (h-swish) are added after constructing the MCNN model to eliminate the potential accuracy loss in PD pattern recognition. The MCNN model is trained and tested with the pre-processed PRPD spectrum, and a variety of methods are used to visualize the model to verify the effectiveness of the model. Finally, the performance of MCNN is compared with many existing PD pattern recognition models based on convolutional neural network (CNN), the results show that the proposed MCNN can further reduce the number of parameters of the model and improve the calculation speed to achieve the best performance on the premise of good recognition accuracy.

A Robust Detection Method for Multilane Lines in Complex Traffic Scenes

“…e results show that, in terms of accuracy, using standard convolution is 1.1% higher than using depth separable convolution, but in terms of calculation amount and parameter amount, the former is 8-9 times the latter. From [20], we can conclude that the use of deep separable convolution can greatly reduce the amount of calculation and the number of parameters while basically ensuring the accuracy. Correspondingly, it can reduce the difficulty of network model training, reduce training time, and reduce the performance requirements of hardware devices.…”

“…In order to compare the performance of convolutional neural networks using standard convolution and depthwise separable convolution, these two networks were used for training and testing on the ImageNet dataset in [20]. e results show that, in terms of accuracy, using standard convolution is 1.1% higher than using depth separable convolution, but in terms of calculation amount and parameter amount, the former is 8-9 times the latter.…”

“…(1) e multilane line detection is considered as an instance segmentation problem in this paper, and a deep learning network is designed to solve the problems of uncertain number of lanes and lane change. e lane detection problem is decomposed into two tasks, that is, lane line semantic segmentation branch and lane line Id embedding branch, and a multitask network based on MobileNet [20] is designed to shorten the forward propagation calculation time without loss of detection accuracy. In MobileNet, a convolution method called depthwise separable convolution is used to replace traditional convolution methods to reduce the network weight parameters.…”

Deep Learning in High Voltage Engineering: A Literature Review

“…The PRPD data of different PD sources in a transformer were collected in a laboratorycontrolled setup and reported in [18]. The PD sources included tip discharge, surface discharge, air gap discharge and suspended discharge.…”

Surface Defect Segmentation Algorithm of Steel Plate Based on Geometric Median Filter Pruning

“…To randomly disrupt the feature channels, ShuffleNet (Xin et al, 2021) divides the feature channels into multiple groups and The pruned model has some robustness and can achieve better optimization convolves them to increase the information exchange between different feature channels. MobileNet (Sun et al, 2021a) designs a deeply separable convolution module and fuses the information of different feature channels by 1 × 1 convolution. In addition, researchers often introduce 1 × 1 filters between 3 × 3 filters to reduce the number of input and output channels of the feature map.…”